A cyber-physical system for quality assessment in welding electric vehicle batteries

• Main Author: Σαμπατακάκης, Κυριάκος
• Other Authors: Sabatakakis, Kyriakos
• Language: English
• Published: 2022
• Subjects: Battery assembly; Laser welding; Quality assessment; Machine learning; Cyber-physical system; Συναρμολόγηση μπαταριών; Συγκόλληση laser; Μηχανική μάθηση; Αξιολόγηση ποιότητας; Κυβερνοφυσικό σύστημα

The battery of an electric vehicle consists of cells assembled into modules which in turn are assembled to create a battery pack. Most of these connections are permanent and are made using a welding process. However, due to the involvement of non-ferrous, inhomogeneous and multilayer materials, the creation of welds with low electrical resistance, high strength and quality characteristics that can be reproduced with minimal deviation is a challenge. This fact combined with the assembly process, in which the interconnection of tens or even hundreds of cells take place, makes the existence even of a single joint that is out-of-spec, crucial for the safe and efficient operation of the battery, but also for its longevity. This fact makes quality inspection of every joint necessary, which cannot be achieved by using sample-based destructive or non-destructive methods. And while the scientific community has developed and described solutions and approaches for non-destructive real-time inspection and assessment in welding a very small amount of them concerns battery assembling applications and in general the welding of non-ferrous dissimilar metals. On the other hand, none of these studies yet has addressed the issue of inspecting or assessing the electrical quality of the joints. In this study based on the design practices of Cyber-Physical systems a novel quality assessment approach for the assembly of batteries in terms of electrical and mechanical quality is developed in the context of the Laser welding of aluminum and copper battery tabs. The assessment approach is based on infrared vision data using machine learning. The classification accuracy for the electrical and mechanical quality of the welds on the training and test data was set at 100% under specific conditions.